Connect With Us

• Website (https://www.rsquaredacademy.com/)
• Free Online R Courses (https://rsquared-academy.thinkific.com/)
• R Packages (https://pkgs.rsquaredacademy.com)
• Shiny Apps (https://apps.rsquaredacademy.com)
• Blog (https://blog.rsquaredacademy.com)
• GitHub (https://github.com/rsquaredacademy)
• YouTube (https://www.youtube.com/user/rsquaredin/)
• Twitter (https://twitter.com/rsquaredacademy)
• Facebook (https://www.facebook.com/rsquaredacademy/)
• Linkedin (https://in.linkedin.com/company/rsquared-academy)

• What?
• Why?
• How?
• Use Cases
• Demo

Introduction

What?

Why ?

Advantages

Use Cases

Simple Example

Steps

Itemset

Antecedent & Consequent

Support

Confidence

Lift

Data

• UCI
• data.world

Data Dictionary

• invoice number
• description
• quantity
• invoice date
• unit price
• customer id
• country

Libraries

library(readxl)
library(readr)
library(mbar)
library(arules)
library(arulesViz)
library(magrittr)
library(dplyr)
library(lubridate)
library(forcats)
library(ggplot2)

Preprocessing

Preprocessing

mba_data     <- read_excel("online-retail.xlsx")
transactions <- mbar_prep_data(mba_data, InvoiceNo, Description)
head(transactions)

What time of day do people purchase?

How many items are purchased on an average?

items <- 
  mba_data %>%
  group_by(InvoiceNo) %>%
  summarize(count = n()) %>%
  pull(count) 

mean(items)
median(items)

Most Purchased Items

mba_data %>%
  group_by(Description) %>%
  summarize(count = n()) %>%
  arrange(desc(count))

Average Order Value

total_revenue <- 
  mba_data %>%
  group_by(InvoiceNo) %>%
  summarize(order_sum = sum(UnitPrice)) %>%
  pull(order_sum) %>%
  sum()

total_transactions <- 
  mba_data %>%
  group_by(InvoiceNo) %>%
  summarize(n()) %>%
  nrow()

total_revenue / total_transactions

Read Data

basket_data <- read.transactions("transaction_data.csv", format = "basket", 
  sep = ",")
basket_data

Summary

summary(basket_data)

Item Frequency Plot

itemFrequencyPlot(basket_data, topN = 10, type = 'absolute')

Generate Rules

rules <- apriori(basket_data, parameter = list(supp=0.009, conf=0.8, 
    target = "rules", maxlen = 4))

Rules Summary

summary(rules)

Inspect Rules

basket_rules <- sort(rules, by = 'confidence', decreasing = TRUE)
inspect(basket_rules[1:10])

Redundant Rules

rules_redundant <-
  basket_rules %>%
  is.subset(rules) %>%
  colSums() %>%
  is_greater_than(1) %>%
  which()

Redundant Rules

rules_redundant

Remove Redundant Rules

rules_pruned <- basket_rules[-rules_redundant]
inspect(rules_pruned)

What influenced purchase of product X?

sugar_rules <- apriori(basket_data, parameter = list(supp = 0.009, conf = 0.8), 
  appearance = list(default = "lhs", rhs = "SUGAR"), 
  control = list(verbose = F)) 
rules_sugar <- sort(sugar_rules, by = "confidence", decreasing = TRUE)
inspect(rules_sugar)

What purchases did product X influence?

sugar_rules <- apriori(basket_data, parameter = list(supp = 0.009, conf = 0.8), 
  appearance = list(default = "rhs", lhs = "SUGAR"), 
  control = list(verbose = F)) 
rules_sugar <- sort(sugar_rules, by = "confidence", decreasing = TRUE)
inspect(rules_sugar)

Top Rules

top_rules <- basket_rules[1:10]
inspect(top_rules)

Scatter Plot

plot(basket_rules)

Network Plot

plot(top_rules, method = 'graph')

Things to keep in mind..

Things to keep in mind..

Summary

• unsupervised data mining technique
• uncovers products frequently bought together
• creates if-then scenario rules
• cost-effective, insightful and actionable
• association rule mining has applications in several industries
• directionality of rule is lost while using lift
• confidence as a measure can be misleading